ISCA Archive - A real time speech decoder using instantaneous frequency and energy (original) (raw)

A real time speech decoder using instantaneous frequency and energy

A. Tsopanoglou, E. D. Kyriakis-Bitzaros, J. Mourjopoulos, George K. Kokkinakis

A real-time speech Acoustic-to-Phonetic Decoder (APD) is presented, suitable for efficient segmentation of speech into phonemes and classification of these phonemes into broad phonetic categories. The system can be implemented in DSP modules (e. g. TMS32025), mainly because the analysis parameters are estimated in the time domain. These parameters are the Envelope Function and the Instantaneous Frequency of the signal evaluated efficiently via digital filters. The system can achieve satisfactory performance, which is both speaker and language independent.